xref: /linux/arch/um/os-Linux/skas/process.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
1 /*
2  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3  * Licensed under the GPL
4  */
5 
6 #include <stdlib.h>
7 #include <unistd.h>
8 #include <sched.h>
9 #include <errno.h>
10 #include <string.h>
11 #include <sys/mman.h>
12 #include <sys/wait.h>
13 #include <asm/unistd.h>
14 #include <as-layout.h>
15 #include <init.h>
16 #include <kern_util.h>
17 #include <mem.h>
18 #include <os.h>
19 #include <proc_mm.h>
20 #include <ptrace_user.h>
21 #include <registers.h>
22 #include <skas.h>
23 #include <skas_ptrace.h>
24 #include <sysdep/stub.h>
25 
26 int is_skas_winch(int pid, int fd, void *data)
27 {
28 	return pid == getpgrp();
29 }
30 
31 static int ptrace_dump_regs(int pid)
32 {
33 	unsigned long regs[MAX_REG_NR];
34 	int i;
35 
36 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
37 		return -errno;
38 
39 	printk(UM_KERN_ERR "Stub registers -\n");
40 	for (i = 0; i < ARRAY_SIZE(regs); i++)
41 		printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
42 
43 	return 0;
44 }
45 
46 /*
47  * Signals that are OK to receive in the stub - we'll just continue it.
48  * SIGWINCH will happen when UML is inside a detached screen.
49  */
50 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
51 
52 /* Signals that the stub will finish with - anything else is an error */
53 #define STUB_DONE_MASK (1 << SIGTRAP)
54 
55 void wait_stub_done(int pid)
56 {
57 	int n, status, err, bad_stop = 0;
58 
59 	while (1) {
60 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
61 		if ((n < 0) || !WIFSTOPPED(status))
62 			goto bad_wait;
63 
64 		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
65 			break;
66 
67 		err = ptrace(PTRACE_CONT, pid, 0, 0);
68 		if (err) {
69 			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
70 			       "errno = %d\n", errno);
71 			fatal_sigsegv();
72 		}
73 	}
74 
75 	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
76 		return;
77 	else
78 		bad_stop = 1;
79 
80 bad_wait:
81 	err = ptrace_dump_regs(pid);
82 	if (err)
83 		printk(UM_KERN_ERR "Failed to get registers from stub, "
84 		       "errno = %d\n", -err);
85 	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
86 	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
87 	       status);
88 	if (bad_stop)
89 		kill(pid, SIGKILL);
90 	else
91 		fatal_sigsegv();
92 }
93 
94 extern unsigned long current_stub_stack(void);
95 
96 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
97 {
98 	int err;
99 
100 	if (ptrace_faultinfo) {
101 		err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
102 		if (err) {
103 			printk(UM_KERN_ERR "get_skas_faultinfo - "
104 			       "PTRACE_FAULTINFO failed, errno = %d\n", errno);
105 			fatal_sigsegv();
106 		}
107 
108 		/* Special handling for i386, which has different structs */
109 		if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
110 			memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
111 			       sizeof(struct faultinfo) -
112 			       sizeof(struct ptrace_faultinfo));
113 	}
114 	else {
115 		unsigned long fpregs[FP_SIZE];
116 
117 		err = get_fp_registers(pid, fpregs);
118 		if (err < 0) {
119 			printk(UM_KERN_ERR "save_fp_registers returned %d\n",
120 			       err);
121 			fatal_sigsegv();
122 		}
123 		err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
124 		if (err) {
125 			printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
126 			       "errno = %d\n", pid, errno);
127 			fatal_sigsegv();
128 		}
129 		wait_stub_done(pid);
130 
131 		/*
132 		 * faultinfo is prepared by the stub-segv-handler at start of
133 		 * the stub stack page. We just have to copy it.
134 		 */
135 		memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
136 
137 		err = put_fp_registers(pid, fpregs);
138 		if (err < 0) {
139 			printk(UM_KERN_ERR "put_fp_registers returned %d\n",
140 			       err);
141 			fatal_sigsegv();
142 		}
143 	}
144 }
145 
146 static void handle_segv(int pid, struct uml_pt_regs * regs)
147 {
148 	get_skas_faultinfo(pid, &regs->faultinfo);
149 	segv(regs->faultinfo, 0, 1, NULL);
150 }
151 
152 /*
153  * To use the same value of using_sysemu as the caller, ask it that value
154  * (in local_using_sysemu
155  */
156 static void handle_trap(int pid, struct uml_pt_regs *regs,
157 			int local_using_sysemu)
158 {
159 	int err, status;
160 
161 	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
162 		fatal_sigsegv();
163 
164 	/* Mark this as a syscall */
165 	UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
166 
167 	if (!local_using_sysemu)
168 	{
169 		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
170 			     __NR_getpid);
171 		if (err < 0) {
172 			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
173 			       "failed, errno = %d\n", errno);
174 			fatal_sigsegv();
175 		}
176 
177 		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
178 		if (err < 0) {
179 			printk(UM_KERN_ERR "handle_trap - continuing to end of "
180 			       "syscall failed, errno = %d\n", errno);
181 			fatal_sigsegv();
182 		}
183 
184 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
185 		if ((err < 0) || !WIFSTOPPED(status) ||
186 		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
187 			err = ptrace_dump_regs(pid);
188 			if (err)
189 				printk(UM_KERN_ERR "Failed to get registers "
190 				       "from process, errno = %d\n", -err);
191 			printk(UM_KERN_ERR "handle_trap - failed to wait at "
192 			       "end of syscall, errno = %d, status = %d\n",
193 			       errno, status);
194 			fatal_sigsegv();
195 		}
196 	}
197 
198 	handle_syscall(regs);
199 }
200 
201 extern int __syscall_stub_start;
202 
203 static int userspace_tramp(void *stack)
204 {
205 	void *addr;
206 	int err;
207 
208 	ptrace(PTRACE_TRACEME, 0, 0, 0);
209 
210 	signal(SIGTERM, SIG_DFL);
211 	signal(SIGWINCH, SIG_IGN);
212 	err = set_interval();
213 	if (err) {
214 		printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
215 		       "errno = %d\n", err);
216 		exit(1);
217 	}
218 
219 	if (!proc_mm) {
220 		/*
221 		 * This has a pte, but it can't be mapped in with the usual
222 		 * tlb_flush mechanism because this is part of that mechanism
223 		 */
224 		int fd;
225 		unsigned long long offset;
226 		fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
227 		addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
228 			      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
229 		if (addr == MAP_FAILED) {
230 			printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
231 			       "errno = %d\n", STUB_CODE, errno);
232 			exit(1);
233 		}
234 
235 		if (stack != NULL) {
236 			fd = phys_mapping(to_phys(stack), &offset);
237 			addr = mmap((void *) STUB_DATA,
238 				    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
239 				    MAP_FIXED | MAP_SHARED, fd, offset);
240 			if (addr == MAP_FAILED) {
241 				printk(UM_KERN_ERR "mapping segfault stack "
242 				       "at 0x%lx failed, errno = %d\n",
243 				       STUB_DATA, errno);
244 				exit(1);
245 			}
246 		}
247 	}
248 	if (!ptrace_faultinfo && (stack != NULL)) {
249 		struct sigaction sa;
250 
251 		unsigned long v = STUB_CODE +
252 				  (unsigned long) stub_segv_handler -
253 				  (unsigned long) &__syscall_stub_start;
254 
255 		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
256 		sigemptyset(&sa.sa_mask);
257 		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
258 		sa.sa_sigaction = (void *) v;
259 		sa.sa_restorer = NULL;
260 		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
261 			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
262 			       "handler failed - errno = %d\n", errno);
263 			exit(1);
264 		}
265 	}
266 
267 	kill(os_getpid(), SIGSTOP);
268 	return 0;
269 }
270 
271 /* Each element set once, and only accessed by a single processor anyway */
272 #undef NR_CPUS
273 #define NR_CPUS 1
274 int userspace_pid[NR_CPUS];
275 
276 int start_userspace(unsigned long stub_stack)
277 {
278 	void *stack;
279 	unsigned long sp;
280 	int pid, status, n, flags, err;
281 
282 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
283 		     PROT_READ | PROT_WRITE | PROT_EXEC,
284 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
285 	if (stack == MAP_FAILED) {
286 		err = -errno;
287 		printk(UM_KERN_ERR "start_userspace : mmap failed, "
288 		       "errno = %d\n", errno);
289 		return err;
290 	}
291 
292 	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
293 
294 	flags = CLONE_FILES;
295 	if (proc_mm)
296 		flags |= CLONE_VM;
297 	else
298 		flags |= SIGCHLD;
299 
300 	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
301 	if (pid < 0) {
302 		err = -errno;
303 		printk(UM_KERN_ERR "start_userspace : clone failed, "
304 		       "errno = %d\n", errno);
305 		return err;
306 	}
307 
308 	do {
309 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
310 		if (n < 0) {
311 			err = -errno;
312 			printk(UM_KERN_ERR "start_userspace : wait failed, "
313 			       "errno = %d\n", errno);
314 			goto out_kill;
315 		}
316 	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
317 
318 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
319 		err = -EINVAL;
320 		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
321 		       "status = %d\n", status);
322 		goto out_kill;
323 	}
324 
325 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
326 		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
327 		err = -errno;
328 		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
329 		       "failed, errno = %d\n", errno);
330 		goto out_kill;
331 	}
332 
333 	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
334 		err = -errno;
335 		printk(UM_KERN_ERR "start_userspace : munmap failed, "
336 		       "errno = %d\n", errno);
337 		goto out_kill;
338 	}
339 
340 	return pid;
341 
342  out_kill:
343 	os_kill_ptraced_process(pid, 1);
344 	return err;
345 }
346 
347 void userspace(struct uml_pt_regs *regs)
348 {
349 	struct itimerval timer;
350 	unsigned long long nsecs, now;
351 	int err, status, op, pid = userspace_pid[0];
352 	/* To prevent races if using_sysemu changes under us.*/
353 	int local_using_sysemu;
354 	siginfo_t si;
355 
356 	/* Handle any immediate reschedules or signals */
357 	interrupt_end();
358 
359 	if (getitimer(ITIMER_VIRTUAL, &timer))
360 		printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
361 	nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
362 		timer.it_value.tv_usec * UM_NSEC_PER_USEC;
363 	nsecs += os_nsecs();
364 
365 	while (1) {
366 		/*
367 		 * This can legitimately fail if the process loads a
368 		 * bogus value into a segment register.  It will
369 		 * segfault and PTRACE_GETREGS will read that value
370 		 * out of the process.  However, PTRACE_SETREGS will
371 		 * fail.  In this case, there is nothing to do but
372 		 * just kill the process.
373 		 */
374 		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
375 			fatal_sigsegv();
376 
377 		if (put_fp_registers(pid, regs->fp))
378 			fatal_sigsegv();
379 
380 		/* Now we set local_using_sysemu to be used for one loop */
381 		local_using_sysemu = get_using_sysemu();
382 
383 		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
384 					     singlestepping(NULL));
385 
386 		if (ptrace(op, pid, 0, 0)) {
387 			printk(UM_KERN_ERR "userspace - ptrace continue "
388 			       "failed, op = %d, errno = %d\n", op, errno);
389 			fatal_sigsegv();
390 		}
391 
392 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
393 		if (err < 0) {
394 			printk(UM_KERN_ERR "userspace - wait failed, "
395 			       "errno = %d\n", errno);
396 			fatal_sigsegv();
397 		}
398 
399 		regs->is_user = 1;
400 		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
401 			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
402 			       "errno = %d\n", errno);
403 			fatal_sigsegv();
404 		}
405 
406 		if (get_fp_registers(pid, regs->fp)) {
407 			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
408 			       "errno = %d\n", errno);
409 			fatal_sigsegv();
410 		}
411 
412 		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
413 
414 		if (WIFSTOPPED(status)) {
415 			int sig = WSTOPSIG(status);
416 
417 			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
418 
419 			switch (sig) {
420 			case SIGSEGV:
421 				if (PTRACE_FULL_FAULTINFO ||
422 				    !ptrace_faultinfo) {
423 					get_skas_faultinfo(pid,
424 							   &regs->faultinfo);
425 					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
426 							     regs);
427 				}
428 				else handle_segv(pid, regs);
429 				break;
430 			case SIGTRAP + 0x80:
431 			        handle_trap(pid, regs, local_using_sysemu);
432 				break;
433 			case SIGTRAP:
434 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
435 				break;
436 			case SIGVTALRM:
437 				now = os_nsecs();
438 				if (now < nsecs)
439 					break;
440 				block_signals();
441 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
442 				unblock_signals();
443 				nsecs = timer.it_value.tv_sec *
444 					UM_NSEC_PER_SEC +
445 					timer.it_value.tv_usec *
446 					UM_NSEC_PER_USEC;
447 				nsecs += os_nsecs();
448 				break;
449 			case SIGIO:
450 			case SIGILL:
451 			case SIGBUS:
452 			case SIGFPE:
453 			case SIGWINCH:
454 				block_signals();
455 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
456 				unblock_signals();
457 				break;
458 			default:
459 				printk(UM_KERN_ERR "userspace - child stopped "
460 				       "with signal %d\n", sig);
461 				fatal_sigsegv();
462 			}
463 			pid = userspace_pid[0];
464 			interrupt_end();
465 
466 			/* Avoid -ERESTARTSYS handling in host */
467 			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
468 				PT_SYSCALL_NR(regs->gp) = -1;
469 		}
470 	}
471 }
472 
473 static unsigned long thread_regs[MAX_REG_NR];
474 static unsigned long thread_fp_regs[FP_SIZE];
475 
476 static int __init init_thread_regs(void)
477 {
478 	get_safe_registers(thread_regs, thread_fp_regs);
479 	/* Set parent's instruction pointer to start of clone-stub */
480 	thread_regs[REGS_IP_INDEX] = STUB_CODE +
481 				(unsigned long) stub_clone_handler -
482 				(unsigned long) &__syscall_stub_start;
483 	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
484 		sizeof(void *);
485 #ifdef __SIGNAL_FRAMESIZE
486 	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
487 #endif
488 	return 0;
489 }
490 
491 __initcall(init_thread_regs);
492 
493 int copy_context_skas0(unsigned long new_stack, int pid)
494 {
495 	struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
496 	int err;
497 	unsigned long current_stack = current_stub_stack();
498 	struct stub_data *data = (struct stub_data *) current_stack;
499 	struct stub_data *child_data = (struct stub_data *) new_stack;
500 	unsigned long long new_offset;
501 	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
502 
503 	/*
504 	 * prepare offset and fd of child's stack as argument for parent's
505 	 * and child's mmap2 calls
506 	 */
507 	*data = ((struct stub_data) { .offset	= MMAP_OFFSET(new_offset),
508 				      .fd	= new_fd,
509 				      .timer    = ((struct itimerval)
510 					           { .it_value = tv,
511 						     .it_interval = tv }) });
512 
513 	err = ptrace_setregs(pid, thread_regs);
514 	if (err < 0) {
515 		err = -errno;
516 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
517 		       "failed, pid = %d, errno = %d\n", pid, -err);
518 		return err;
519 	}
520 
521 	err = put_fp_registers(pid, thread_fp_regs);
522 	if (err < 0) {
523 		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
524 		       "failed, pid = %d, err = %d\n", pid, err);
525 		return err;
526 	}
527 
528 	/* set a well known return code for detection of child write failure */
529 	child_data->err = 12345678;
530 
531 	/*
532 	 * Wait, until parent has finished its work: read child's pid from
533 	 * parent's stack, and check, if bad result.
534 	 */
535 	err = ptrace(PTRACE_CONT, pid, 0, 0);
536 	if (err) {
537 		err = -errno;
538 		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
539 		       "errno = %d\n", pid, errno);
540 		return err;
541 	}
542 
543 	wait_stub_done(pid);
544 
545 	pid = data->err;
546 	if (pid < 0) {
547 		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
548 		       "error %d\n", -pid);
549 		return pid;
550 	}
551 
552 	/*
553 	 * Wait, until child has finished too: read child's result from
554 	 * child's stack and check it.
555 	 */
556 	wait_stub_done(pid);
557 	if (child_data->err != STUB_DATA) {
558 		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
559 		       "error %ld\n", child_data->err);
560 		err = child_data->err;
561 		goto out_kill;
562 	}
563 
564 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
565 		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
566 		err = -errno;
567 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
568 		       "failed, errno = %d\n", errno);
569 		goto out_kill;
570 	}
571 
572 	return pid;
573 
574  out_kill:
575 	os_kill_ptraced_process(pid, 1);
576 	return err;
577 }
578 
579 /*
580  * This is used only, if stub pages are needed, while proc_mm is
581  * available. Opening /proc/mm creates a new mm_context, which lacks
582  * the stub-pages. Thus, we map them using /proc/mm-fd
583  */
584 int map_stub_pages(int fd, unsigned long code, unsigned long data,
585 		   unsigned long stack)
586 {
587 	struct proc_mm_op mmop;
588 	int n;
589 	unsigned long long code_offset;
590 	int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
591 				   &code_offset);
592 
593 	mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
594 				      .u         =
595 				      { .mmap    =
596 					{ .addr    = code,
597 					  .len     = UM_KERN_PAGE_SIZE,
598 					  .prot    = PROT_EXEC,
599 					  .flags   = MAP_FIXED | MAP_PRIVATE,
600 					  .fd      = code_fd,
601 					  .offset  = code_offset
602 	} } });
603 	CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
604 	if (n != sizeof(mmop)) {
605 		n = errno;
606 		printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
607 		       "offset = %llx\n", code, code_fd,
608 		       (unsigned long long) code_offset);
609 		printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
610 		       "failed, err = %d\n", n);
611 		return -n;
612 	}
613 
614 	if (stack) {
615 		unsigned long long map_offset;
616 		int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
617 		mmop = ((struct proc_mm_op)
618 				{ .op        = MM_MMAP,
619 				  .u         =
620 				  { .mmap    =
621 				    { .addr    = data,
622 				      .len     = UM_KERN_PAGE_SIZE,
623 				      .prot    = PROT_READ | PROT_WRITE,
624 				      .flags   = MAP_FIXED | MAP_SHARED,
625 				      .fd      = map_fd,
626 				      .offset  = map_offset
627 		} } });
628 		CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
629 		if (n != sizeof(mmop)) {
630 			n = errno;
631 			printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
632 			       "data failed, err = %d\n", n);
633 			return -n;
634 		}
635 	}
636 
637 	return 0;
638 }
639 
640 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
641 {
642 	(*buf)[0].JB_IP = (unsigned long) handler;
643 	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
644 		sizeof(void *);
645 }
646 
647 #define INIT_JMP_NEW_THREAD 0
648 #define INIT_JMP_CALLBACK 1
649 #define INIT_JMP_HALT 2
650 #define INIT_JMP_REBOOT 3
651 
652 void switch_threads(jmp_buf *me, jmp_buf *you)
653 {
654 	if (UML_SETJMP(me) == 0)
655 		UML_LONGJMP(you, 1);
656 }
657 
658 static jmp_buf initial_jmpbuf;
659 
660 /* XXX Make these percpu */
661 static void (*cb_proc)(void *arg);
662 static void *cb_arg;
663 static jmp_buf *cb_back;
664 
665 int start_idle_thread(void *stack, jmp_buf *switch_buf)
666 {
667 	int n;
668 
669 	set_handler(SIGWINCH);
670 
671 	/*
672 	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
673 	 * and restore signals, with the possible side-effect of
674 	 * trying to handle any signals which came when they were
675 	 * blocked, which can't be done on this stack.
676 	 * Signals must be blocked when jumping back here and restored
677 	 * after returning to the jumper.
678 	 */
679 	n = setjmp(initial_jmpbuf);
680 	switch (n) {
681 	case INIT_JMP_NEW_THREAD:
682 		(*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
683 		(*switch_buf)[0].JB_SP = (unsigned long) stack +
684 			UM_THREAD_SIZE - sizeof(void *);
685 		break;
686 	case INIT_JMP_CALLBACK:
687 		(*cb_proc)(cb_arg);
688 		longjmp(*cb_back, 1);
689 		break;
690 	case INIT_JMP_HALT:
691 		kmalloc_ok = 0;
692 		return 0;
693 	case INIT_JMP_REBOOT:
694 		kmalloc_ok = 0;
695 		return 1;
696 	default:
697 		printk(UM_KERN_ERR "Bad sigsetjmp return in "
698 		       "start_idle_thread - %d\n", n);
699 		fatal_sigsegv();
700 	}
701 	longjmp(*switch_buf, 1);
702 }
703 
704 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
705 {
706 	jmp_buf here;
707 
708 	cb_proc = proc;
709 	cb_arg = arg;
710 	cb_back = &here;
711 
712 	block_signals();
713 	if (UML_SETJMP(&here) == 0)
714 		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
715 	unblock_signals();
716 
717 	cb_proc = NULL;
718 	cb_arg = NULL;
719 	cb_back = NULL;
720 }
721 
722 void halt_skas(void)
723 {
724 	block_signals();
725 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
726 }
727 
728 void reboot_skas(void)
729 {
730 	block_signals();
731 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
732 }
733 
734 void __switch_mm(struct mm_id *mm_idp)
735 {
736 	int err;
737 
738 	/* FIXME: need cpu pid in __switch_mm */
739 	if (proc_mm) {
740 		err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
741 			     mm_idp->u.mm_fd);
742 		if (err) {
743 			printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
744 			       "failed, errno = %d\n", errno);
745 			fatal_sigsegv();
746 		}
747 	}
748 	else userspace_pid[0] = mm_idp->u.pid;
749 }
750